Academic Radar

The Bring-Your-Own-Key (BYOK) paradigm in real-world agents lets users route LLM traffic through third-party relay services while authenticating with their own credentials. We identify a fundamental security gap in this architecture, which we call *post-alignment tampering*: an a...

Malicious LLM outputs in this domain pose hardware-level threats, including hardware Trojan insertion, side-channel leakage, and intellectual property theft, that are irreversible once fabricated into silicon. We present the HarmChip benchmark to assess jailbreak susceptibility i...

That said, they may detect the security vulnerabilities of traditional workflows or jailbreak an LLM or an LLM agent, but the security of agentic workflows is...

# LITMUS: Benchmarking Behavioral Jailbreaks of LLM Agents in Real OS Environments. The rapid proliferation of LLM-based autonomous agents in real operating system. environments introduces a qualitatively new category of safety risk beyond. induces an agent to execute dangerous O...

# [Uncaptioned image]     The Art of the Jailbreak: Formulating Jailbreak Attacks for LLM Security Beyond Binary Scoring. Unlike prior datasets that treat jailbreaks as undifferentiated collections, every prompt in our dataset is assigned to one of 14 cybersecurity attack categor...

# Guaranteed Jailbreaking Defense via Disrupt-and-Rectify Smoothing. This paper proposes a guaranteed defense method for large language models (LLMs) to safeguard against jailbreaking attacks. Drawing inspiration from the denoised-smoothing approach in the adversarial defense dom...

Many-shot jailbreaking (MSJ) causes safety-aligned language models to answer harmful queries by preceding them with many harmful question-answer demonstrations. We study why this attack becomes stronger as the number of demonstrations increases. Empirically, we find that MSJ indu...

We propose *Prompt Embedding Optimization* (PEO), a multi-round white-box jailbreak that directly optimizes the embeddings of the original prompt tokens without appending any adversarial tokens, and show that the concern is unfounded: the optimized embeddings remain close enough ...

Zhang and Luo (2024) proposed a psychotherapy-oriented LLM framework that incorporates a dual-memory ... reinforcement learning with human feedback (RLHF) to...

# PERSA: Reinforcement Learning for Professor-Style Personalized Feedback with LLMs. Large language models (LLMs) can provide automated feedback in educational settings, but aligning an LLM’s style with a specific instructor’s tone while maintaining diagnostic correctness remains...

We present MemBuilder, a framework for training models to construct long-term memory with attributed dense rewards. Our architecture utilizes a multi-...

In reinforcement learning from human feedback (RLHF), human preferences over model outputs are used to train a reward model that assigns scalar values to responses. Because these rewards are inferred from pairwise comparisons, this learning depends on an assumed relationship betw...

# The Landscape of Agentic Reinforcement Learning for LLMs: A Survey. The emergence of agentic reinforcement learning (Agentic RL) marks a paradigm shift from conventional reinforcement learning applied to large language models (LLM RL), reframing LLMs from passive sequence gener...

# Reinforcement Learning for LLM-based Multi-Agent Systems through Orchestration Traces. As large language model (LLM) agents evolve from isolated tool users. into coordinated teams, reinforcement learning (RL) must optimize not. This paper studies RL for. LLM-based multi-agent s...

# StepPO: Step-Aligned Policy Optimization for Agentic Reinforcement Learning. Agentic Reinforcement Learning (RL) is emerging as a central post-training paradigm for empowering LLMs with these capabilities and is playing an increasingly pivotal role in agent training. Unlike sin...

Reinforcement learning is a standard post-training mechanism for improving the reasoning capabilities of large language models, yet performance remains sensitive to the design of the reward function that drives policy optimization. This paper introduces a search-driven framework ...

# Computer Science > Distributed, Parallel, and Cluster Computing. # Title:ZipCCL: Efficient Lossless Data Compression of Communication Collectives for Accelerating LLM Training. | Subjects: | Distributed, Parallel, and Cluster Computing (cs.DC); Computation and Language (cs.CL) ...

# Computer Science > Distributed, Parallel, and Cluster Computing. # Title:TACO: Efficient Communication Compression of Intermediate Tensors for Scalable Tensor-Parallel LLM Training. | Subjects: | Distributed, Parallel, and Cluster Computing (cs.DC); Artificial Intelligence (cs....

Distributed training scales LLMs beyond single-device limits through complementary parallelization strategies. Data Parallelism (DP). DP replicates model states...

TokenWeave: Efficient Compute-Communication Overlap for Distributed LLM Inference. Distributed inference of large language models (LLMs) using tensor parallelism can introduce communication overheads of 2020% even over GPUs connected via NVLink, a high-speed GPU interconnect. Thi...

# Federation of Experts: Communication Efficient Distributed Inference for Large Language Models. In a single-node setting, FoE completely eliminates all-to-all communication as all experts within a group are contained on the same GPU. In multi-node settings, FoE confines all-to-...

# ZipCCL: Efficient Lossless Data Compression of Communication Collectives for Accelerating LLM Training. We observe that the communication data, including activations, gradients and parameters, during training often follows a near‑Gaussian distribution, which is a key feature fo...

We propose a pipeline that adapts a text-to-speech (TTS) decoder to a target-accent speaker using fewer than ten reference utterances and employs large language...

# OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models. We present OmniVoice, a massively multilingual zero-shot text-to-speech (TTS) model that scales to over 600 languages. Unlike conventional discrete NAR models that suffer from performance bo...

# VocalParse: Towards Unified and Scalable Singing Voice Transcription with Large Audio Language Models. Despite their utility, current automatic transcription systems face significant challenges: they often rely on complex multi-stage pipelines, struggle to recover text-note ali...

VocalParse: Towards Unified and Scalable Singing Voice Transcription with Large Audio Language Models Yukun Chen1,2 Tianrui Wang3,2 Zhaoxi Mu4,5 Xinyu Yang1∗ EngSiong Chng2∗ 1Xi’an Jiaotong University 2Nanyang Technological University 3Tianjin University 4Ant Group 5Zhejiang Univ...

Few-Shot Accent Synthesis for ASR with LLM-Guided Phoneme Editing Yurii Halychanskyi1,2, Nimet Beyza Bozdag1, Mark Hasegawa-Johnson1, Dilek Hakkani-T¨ ur1, Volodymyr Kindratenko1,2 1 University of Illinois Urbana-Champaign, USA 2 National Center for Supercomputing Applications, U...

Rather, they rely on traces from the vocoding stage, that reconstructs the final waveform from the intermediate acoustic features—i.e., on neural encoding...

# ATRIE: Adaptive Tuning for Robust Inference and Emotion in Persona-Driven Speech Synthesis. High-fidelity character voice synthesis is a cornerstone of immersive multimedia applications, particularly for interacting with anime avatars and digital humans. To bridge this gap, we ...

Abstract. Language model (LM)-based speech enhancement (SE) can generate natural-sounding speech, but under severe noise it often suffers from unreliable...